Keywords: igo, baduk, weiqi, search, trees, games, AI, patterns, hashing, indexing, machine learning, MCTS, UCT, alpha-beta, 9x9, shodan go bet, monte carlo, algorithms, data scientist, data analyst
This is a long-term project (started in 1993 and still far from finished!) to make a program that can play the game of go at the level of a very strong human player. The level of computer go has improved dramatically in the past five years, but there is still much potential for improvement.
This project is a test-bed for my Artificial Intelligence research, in particular data mining, pattern recognition, two-dimensional pattern hashing/indexing and various search algorithms. The necessity for understanding context (in order to make useful patterns) also gives it much in common with my other AI interests of machine translation, automated financial trading and intelligent search.
For the past eight or nine years I have been working on a specific subset of the computer go problem: 9x9. My research started on trying to solve endgames on 9x9 boards (which is a "right-sized" challenge, as it requires strong life and death reading), then moved into using programs together in a team. I presented a peer-reviewed paper entitled A Human-Computer Team Experiment for 9x9 Go at the Computer Games 2010 conference. My research is still continuing, with the eventual aim to be able to state the correct komi for 9x9 go with strong confidence (implying being able to produce a near-perfect opening book).
In 1997 I made a bet with John Tromp that a computer would be able to beat him by the end of 2010, which resulted in the Shodan Go Bet event held in London in December 2010. I ended up losing the bet, and one thousand dollars, but it was good fun and educational, and with hindsight it looks like John may have been cutting it fine and the result might go the other way if the event took place just 12 months later.
As computer go, for me, is mainly a test-bed for more real-world problems I've quite a few half-finished papers that I never seem able to justify the time on finalizing. But a few bits and pieces have been published and can be found at Darren's Computer Go Pages.